Breaking

Grok's Public Account Posts Unrelated Sexual-Assault Reply to Profile-Bio Query

September 18, 2026 at 05:42 EDT

Foundation Models
AI Agents

Grok's public account posted an unrelated first-person sexual-assault statement after a user asked it to restate a profile bio, turning a routine profile-reading prompt into another test of public-facing AI safety. The exchange took place on August 12, 2026: the user prompt was posted at 19:59:34 UTC, and Grok's reply appeared 45 seconds later.

Aug. 12, 2026 · Grok

A Public Reply Went Off Script

A routine profile-bio prompt drew an unrelated first-person sexual-assault response, putting source grounding and public AI behavior back under scrutiny.

45 sec

prompt-to-reply interval

500k

Grok 4.6 context tokens

30+

advertised languages

15 sec

advertised video length

API Price Scale

Grok 4.6 output tokens cost three times the listed input-token price.

input per 1M tokens

output per 1M tokens

Failure Path

Profile-bio request

→

Public AI account

→

Ungrounded sensitive reply

→

Safety scrutiny

Why users want it

Live assistants can summarize profiles, threads, files and events quickly when their answers stay tied to the requested source.

Why it matters

A public, ungrounded answer can turn a small prompt-following miss into a visible trust and safety problem.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.